火灾事故频繁发生，做好火灾受损评估工作，有助于查明起火原因、损失核定、灾后修复等工作顺利开展。针对火灾勘验墙体受损评估中存在的速度慢、结果不够准确、过于依赖个人经验等问题，提出了一种墙体火灾受损程度智能评估方法（YOLOv5-wall damage assessment, YWDA）。该方法基于YOLOv5（you only look once v5）网络进行改进，首先在加强特征提取网络插入坐标注意力（coordinate attention, CA）机制模块，提高了网络对墙体火灾受损特征的检测能力；其次，在损失函数中引入Focal loss，缓解了数据样本不平衡的问题。结合火灾勘验现场全景三维模型，建立一个居民住宅火灾受损评估数据集，并在此数据集上进行实验验证。结果表明：所提方法YWDA相较于其他算法，具有速度快、精度高、模型小等优点，在实际评估任务中具有较强的鲁棒性。因此，所提方法满足墙体火灾受损评估任务高效性、准确性、客观性等要求，可为现代化火灾勘验工作提供技术支持。

Fire accidents are commonplace. Effective fire damage assessment is pivotal for determining the fire's origin, validating losses, and facilitating post-disaster repairs. Addressing challenges in fire investigation, particularly those related to wall damage assessment-like slow inspection speeds, inconsistent results, and heavy reliance on individual experience, an intelligent wall fire damage assessment technique termed YOLOv5-wall damage assessment (YWDA) was proposed. YWDA was based on the you only look once v5 (YOLOv5) network. Primarily, the coordinate attention (CA) module was incorporated into the augmented feature extraction network, which enhancing the detection of wall fire damage characteristics. Additionally, Focal loss was integrated into the loss function to counter the imbalance of data samples. A residential fire damage assessment dataset was built through panoramic 3D modes of the fire investigation site and experimentally verified. The outcomes reveal that the YWDA outperforms alternative algorithms in speed, precision, model compactness, and robustness in real-world evaluations. Consequently, YWDA satisfies the efficiency, accuracy, and objectivity criteria for wall damage assessment, offering valuable technical reinforcement for contemporary fire investigations.